Microsoft launches its first advanced reasoning AI
▼ Summary
– Microsoft announced MAI-Thinking-1, a new flagship in-house AI model, at Build 2026, marking a step away from its previous reliance on OpenAI’s models after renegotiating their deal.
– Microsoft claims MAI-Thinking-1 is a medium-sized model that matches leading models on key software engineering benchmarks and was trained on clean data without distillation from third-party models.
– Other announced models include MAI-Image 2.5 for image generation and editing, and MAI-Transcribe-1.5, which is reportedly five times faster than competing models.
– MAI-Voice-2 and its flash version add 15 new languages and new voice options, with the flash version coming soon.
– The new coding model, MAI-Code-1-Flash, is described as inference-efficient and is integrated into GitHub Copilot and Visual Studio Code.
Microsoft has unveiled its first advanced reasoning AI model during the Microsoft Build 2026 conference, marking a significant shift in the company’s AI strategy. The new “flagship” model, dubbed MAI-Thinking-1, represents a major leap for Microsoft in developing its own AI technology internally. Until last year, the tech giant relied heavily on OpenAI’s models, but the two firms recently renegotiated their partnership to reduce their interdependence.
Described by Microsoft as a “medium-sized model,” MAI-Thinking-1 is said to “match leading models” in critical software engineering benchmarks. The company emphasizes that it “trained it from the ground up on clean data, without distillation from third-party models,” underscoring a commitment to originality and performance.
Beyond the reasoning model, Microsoft introduced several other specialized AI tools at Build 2026. The MAI-Image 2.5 series, including a flash version, handles both text-to-image generation and image editing. For transcription, MAI-Transcribe-1.5 claims to be “five times faster than competing models.” In the voice domain, MAI-Voice-2 and its flash variant (labeled “coming soon”) add support for 15 new languages and expanded voice options. Finally, the new coding model MAI-Code-1-Flash is built for inference efficiency and is integrated directly into GitHub Copilot and Visual Studio Code.
(Source: The Verge)
